Utilisation d'une méthode basée sur l'intelligence artificielle pour modéliser le comportement de barrages en remblai

[Departement_IRSTEA]Eaux [TR1_IRSTEA]RIVAGEInternational audienceDams reliability and safety must be controlled through their whole life to guarantee the safety of the people and assets located at their downstream, to insure they fulfil the functions for which they were built and to safeguard these infrastructures. As a consequence, it is necessary to develop methods and tools for the control of the reliability and safety of dams. In particular, tasks related to the assessment, the diagnosis and the prognosis of dam behaviour are very important to prevent accidents and determine corrective actions. The analysis and control of the dams' behaviour, with the aim of controlling reliability and safety, require the development of models that represent the dynamics of dams. These models have to represent the complexity of the dam behaviour while being easy to handle and to instantiate to the whole set of dams that present different features. They must admit as inputs, miscellaneous data. Finally, these models must be adequate to carry out diagnosis or prognosis tasks to improve dam reliability and safety. In this paper, a multi-model approach is proposed to consider the past, the present and the future behaviour of the dam. Our approach is based on four models: a structural model describing relations between components, a functional model describing the relations which determine the affectation of a possible value to a variable, a behavioural model describing the states of the system and the discrete events that represent the state transitions and a perception model that defines the process and its operating modes. These models were validated by a panel of experts. The method was applied to a real case study concerning a dam which suffered from an internal erosion mechanism. These models were developed to carry out diagnosis, assessment, prognosis and control tasks. Currently, our works show that the developed models are relevant to perform assessment and diagnosis tasks. Moreover, the multi-model approach presents the advantage to facilitate the knowledge representation and handling as well as the results communication to owners or reservoir operators. One of the perspectives is to use the models to forecast dam reliability and safety through tim